OpenAI Announces Open-Weight Model with Reasoning

OpenAI plans to release a language model without usage restrictions for the first time since GPT-2. CEO Sam Altman promises a “very, very good model” with advanced logical reasoning abilities. This marks a strategic shift, as OpenAI had distanced itself from open-source with GPT-3. The term “Open-Weight” suggests that while model weights will be made public, training data and full details will not. Before release, OpenAI will conduct safety tests internally and with external experts, following its Preparedness Framework.